eThSrSSKJrJrJr \"5(aSSKr\R "\5rSr Sr Sr S Sjr S S jr g) z2HQQ (Half-Quadratic Quantization) integration file)is_hqq_availableis_torch_availableloggingNcBUR5H upXlM gN) named_modulesname)modelr modules U/var/www/auris/envauris/lib/python3.13/site-packages/transformers/integrations/hqq.pyautoname_modulesrs++-  .cSRURS5Vs/sH#oS;dM UR5(aM!UPM% sn5$s snf)N.)r layers)joinsplit isnumeric)r ns r name_to_linear_tagr s? 88 3l1>Q5QQ\]\g\g\iQl mmls AAAc[5(aSSKJn [5nUR 5HLup4[ U[ RRW45(dM2UR[U55 MN [U5$)Nr) HQQLinear) rhqq.core.quantizersetr isinstancetorchnnLinearaddrlist)r r linear_tagsr r s r get_linear_tagsr#%sc/%K++-  fuxx : ; ; OO.t4 5.  rcRUR5GHupEUc/nURU5 [U[RR 5(a[ UR5nXa;aVXbQXURUl [U5URUl URURS5 SnSHn[XWS5 M [[UR!555S:a [#UUUS9upUR%S5 GM X4$)NFT)W_qmetar patch_paramshas_been_replaced)named_childrenappendrrrrrr _modules quant_configtype source_clsrequires_grad_setattrlenr!children_prepare_for_hqq_linearpop) r r(r)current_key_namer r linear_tagatt_s r r5r50s,,.   #! % fehhoo . .+FKK8J)+78D8PENN4(56:6lENN4(3NN4(77> $ 'T*' tFOO%& '! +#:)"3$ A R ?/B  ##rc"^Uc/OUn[U5 [U5mURnURn[ [ T5[ U5- [ U5- 5m[ U4SjUR555(a'[RT5nURU5 O[RTU5n[XUS9upUURUS.URlU(d[R!S5 U$)a Prepares nn.Linear layers for HQQ quantization. Since each layer type can have separate quantization parameters, we need to do the following: 1- tag each module with its name via autoname_modules() 2- Extract linear_tags (e.g. ['self_attn.q_proj', ...]) 3- Map quantization parameters as a dictionary linear_tag -> quant_params as HQQLinear expects it, this is referred to as patch_params c3,># UH oT;v M g7fr).0keyr"s r )prepare_for_hqq_linear..ks =)<#+ )YYZK =):):)< ===}}[1 L)}}[,? 6 rSsI5AA   H % n "$J,r